The constraints are not met, but the cost is not infinity

[brunaond]

Hello,

I am using uQuadProg to solve problem with several inequalities and no equalities. It happens, that for some solutions "x" the citeria "CI^T*x+ci0 >= 0" is not met, but the cost function is not equal to infinity. I am passing empty matrices CE and ce0. If I try to pass something like CE=0 ce0 = 0, the program crashes reporting, that the CE and ce0 is linearly dependent.

My understanding is that if the criteria CI^T*x+ci0 >= 0 is not met, then the problem is infeasible and therefore the cost function should be infinity. Am I wrong here? How should I interpret the result x then?

Thanks.

[rcharron]

From the file EigenQP.h
"The function will return the cost of the solution written in the x vector or
std::numeric_limits::infinity() if the problem is infeasible."
So yes it should be infinity.
It would be great if you could give an example where this case happends

[AlanSixth]

@brunaond I think I have run into the same problem here where the solver returns an incorrect solution that disregards inequality constraints after adding the trivial equality constraints. To work around that, I manually set the number of equality constraints to zero.

int p=CE.cols(); p=0;

Also I think this problem may have come all the way since the beginning. See liuq/QuadProgpp#11.